DollhouseMCP 2.0

Building blocks for AI customization.

DollhouseMCP turns prompts into modular building blocks you can create, activate, combine, version, and share: personas, skills, templates, agents, memories, and ensembles.

Get Started on GitHub Install from NPM Browse Collection Dollhouse Research Read Blog

DollhouseMCP is part of Dollhouse Research, the broader platform for user-owned AI customization and control.

Quick Install

One command. Guided install for popular MCP clients.

Copy this command, paste it into a terminal, and run it. DollhouseMCP opens a local web flow that walks you through setup for Claude Desktop, Claude Code, Cursor, VS Code, Codex, Gemini CLI, and other popular MCP clients.

Launch the guided installer

Copy this into a terminal and run it to start the local web installer for the major AI MCP clients.

npx @dollhousemcp/mcp-server@latest --web

The guided flow opens a local web server and helps install DollhouseMCP into the most popular MCP clients.

Prefer to configure things yourself? See the full setup guide.

What Ships With DollhouseMCP

The 2.0 platform is more than personas. It ships a complete element model, a local portfolio, a public collection, MCP-AQL for structured operations, and a safety-aware execution layer for real agent workflows.

Elements

Six reusable element types

Personas, skills, templates, agents, memories, and ensembles are modular building blocks for AI customization across behavior, capabilities, structure, execution, context, and coordinated stacks.

Portfolio

38 bundled starter elements

The server ships with starter personas, skills, templates, agents, memories, and ensembles, including the Dollhouse expert suite and a session monitor agent.

MCP-AQL

CRUDE semantic endpoints

MCP-AQL groups operations into Create, Read, Update, Delete, and Execute, with introspection built in so an LLM can discover what the server supports at runtime.

Permissioning

Dynamic permissioning through active elements

Personas, skills, and ensembles can act as security principals, changing what the AI can do, what needs approval, and what gets denied outright.

Portfolio

Local portfolio with GitHub sync

Your elements live in a readable local portfolio that works offline, can sync to GitHub, and can submit community-ready content back into the collection.

Operations

Full platform operations, not just activation

2.0 includes collection install and submission, portfolio sync, browser opening, enhanced search, skill conversion, execution lifecycle control, and handoff/resume flows.

Security

Autonomy evaluator and danger zone enforcement

Agents receive continue, pause, or escalate guidance every step, while high-risk operations like destructive commands and external calls can be hard-blocked.

Licensing

AGPL core, commercial licensing available

The software is open source under AGPL-3.0-or-later, with commercial licensing available for teams that need proprietary, hosted, or enterprise procurement terms.

How the platform works

Create or edit a Dollhouse element in natural language. Store it in your local portfolio. Activate it when you need it. Combine it with other elements in ensembles. Install community-built elements from the Collection when you want a faster starting point.

Create, edit, or import

Build elements from chat, install them from the Collection, or convert compatible skills into first-class Dollhouse elements.

Keep them in your portfolio

The portfolio lives locally, works offline, and can be backed up or synced to GitHub without giving up file-level ownership.

Activate stacks, not one-off prompts

Mix personas, skills, templates, memories, and ensembles so your AI has reusable behavior instead of re-explaining preferences every session.

Run agents inside a controlled loop

Execution flows back through Gatekeeper, autonomy evaluation, and danger-zone checks so higher agency does not mean less control.

Element model

Personas shape behavior, skills add capabilities, templates structure outputs, agents pursue goals, memories persist context, and ensembles package multiple elements together.

Local portfolio

DollhouseMCP works from a local portfolio, so the things you build stay readable, portable, and under your control before you ever connect GitHub or share publicly.

Collection and sharing

The Dollhouse Collection is the public browse-and-install path for community elements, with install and submission flows built into the platform instead of bolted on later.

Agent execution

Runtime execution stays inside an approval-aware loop with server-side checks, step recording, and a clear path for human input when autonomy should pause.

Why 2.0 matters

MCP-AQL
Five semantic endpoints reduce tool sprawl, improve discoverability, and let the server describe itself at runtime through introspection.
Gatekeeper
Permissions are enforced server-side, so active Dollhouse elements can shape what the AI can do, not just how it sounds.
Portfolio and GitHub workflows
Elements stay local-first with backup, sync, portfolio browser, and collection workflows instead of disappearing into one-off chat history.
Composable execution
Agents, memories, and ensembles turn isolated prompts into reusable workflows that can be inspected, refined, handed off, resumed, and run again.
Licensing clarity
The open-source core stays under AGPL-3.0-or-later, while commercial licensing is available for proprietary or hosted deployment models.

Start with the GitHub repository or the NPM package, then explore the Collection for reusable elements and the blog for deeper technical context.